RoCo-News: A Hand Validated Journalistic Corpus of Romanian

The paper briefly describes the Ro-Co project and, in greater details, one of its first outcomes, the RoCo-News corpus. Ro-Co- is a series of various registers corpora of Romanian that are developed within the Research Institute for Artificial Intelligence of the Romanian Academy. They are planned for public release to the research community as the underlying automatic annotations are validated by human experts. Currently the automatically created data, but not human validated, covers three registers: News, Literature and Legislation and consists of more than 35 millions of lexical tokens. All the corpora of the Ro-Co project are XML annotated and the minimal attributes for each lexical token are its morpho-lexical tag, and lemma. Additionally, the lexical tokens may have specified the hyphenation, a sense identifier and a chunk identifier (specifying to which syntactic chunk the token belongs). In case parallel texts in other languages are available (which, currently, is the case for about 80% of the RoCo corpora) sentence and word alignments are also included. A further extension will be the inclusion of dependency relations among the lexical tokens as well as anaphor-antecedent relations.